The Synthetic Data Generator is an industry first

Chatterbox's Synthetic Data Generator was built to alleviate an industry wide burden that hinders even greater adoption of Artificial Intelligence across the enterprise.

Chatterbox's proprietary Reinforcement Learning & Genetic Algorithm techniques and methods deliver an Enterprise offering that enables partners to manually process only 200 labelled data points as opposed to an industry gold standard acceptance of 1000 per class.

Produce machine learning ready data 80% faster for enterprise outcomes

Learning Process

  • State is the state of the dataset (labelled and unlabelled instances, or simply data points)
  • Actions are RL operations which modify the dataset
    • to label unlabelled data
    • Operations will create a new dataset utilizing technologies such as NNs, GAs, LDA, etc
    • The end result is a fully labelled dataset equivalent to human accuracy
  • The SDG requires at least 20% of an industry expected labelled dataset in order to produce best state and actions with RL methods
  • Within our RL method operations (secret sauce) we form and utilize synthetic instances in order to create the optimal training dataset
Our secret sauce lies within the sequence of operations which create synthetic instances.

Reinforcement Learning Operations

The Synthetic Data Generator

Synthetic Data Generator Market Positioning


Subject Matter Experts (SMEs) / Data Analysts (DAs):
  • SMEs & DAs can accelerate data outcomes up to 80% faster
  • Tackle projects previously not considered or even possible
  • SMEs can create their own data with relevant industry expertise
Enterprise Professional Services:
  • Reduce Cognitive project outcomes from 6 months to 1 week
  • No longer reliant on Offshore Labour arbitrage business models
  • Non-data scientists can prepare Machine Learning ready data in hours
Developers:
  • Accelerate data outcomes by testing, training and iterating in hours
  • Build real world datasets with speed and scale
  • Built using supervised, semi-supervised and unsupervised methods

Reinforcement Learning & Enterprise Adoption

The most common uses of Reinforcement Learning in the industry to date fall into the areas of gaming, agent based simulations and dialogue systems, which are predominantly consumer led interactions. Reinforcement Learning differs from traditional machine learning in that correct examples are never presented to the system for training. Instead exploration or exploitation actions are awarded points based on success (winning) within a particular use case.


Reinforcement Learning Methods

Chatterbox has adopted the Reinforcement Learning Method:

  • SARSA (λ): state, action, reward (next) state, (next) action (therefore the name)
  • On-policy algorithm: policy is updated as soon as the new actions are taken